Nathan Lichtlé

> It seems your compilation scripts does not use the version installed by apt-get. The installed library is `/usr/lib/x86_64-linux-gnu/libsfml-window.so.2.5.1` but you link to `/home/nathan/miniconda3/envs/nocturne/lib/libsfml-window.so.2.5.1` Oh yeah this is likely due...

I will look into that. About version - I just ran `sudo apt-get install libsfml-dev` as mentioned on the SFML tutorial. I'm not an expert, I assumed this automatically installs...

@alex-petrenko Thank you for the answer! That's very appreciated. I am trying to put something together but am still struggling to understand how the code works. So from what you...

Also realizing now that since I barely need any performances for eval, another solution could be to spawn a subprocess (at the start of training) that runs my eval (enjoy)...

In the path, I am using `Gear.FORWARD` or `Gear.BACKWARD` to specify whether the straight line/turn is going forward (positive t, u or v) or backward (negative t, u or v)....

You seem right, I found a route where adding these missing paths was more optimal. Thank you for finding that! Path lengths shouldn't be negative though, according to the paper,...

> Besides, I find that the optimal path length shown in code output is identical with or without the positive condition. (This result seems convincing) @XinArkh I did a quick...

@XinArkh By length, do you mean PathElement.param? If so, seems normal to me that a length remains positive, as the sign is handled through the forwards/backwards gear.

So what you're saying is that if you generate random points (without the missing deepcopy mentioned above), you never get more optimal paths by removing the >0 conditions?

@XinArkh